Laser Diode driver - kHz modulation & 1% duty cycles

[smoothVTer]

Hello Everybody

I've been working with laser diodes most recently and have been able to come up with an almost-working design, albeit with some significant hurdles I have yet to overcome. Basically, I cannot get the transient response I am looking for when PWMing at a few kHz and going to very low duty cycles.   Submitting here to get your expert advice.  First, the schematic:



The theory of operation:   When a current flows through the LD, a reverse current proportional to the light output is generated through the PD.  The LD and PD are in the same 3-pin package ( this being an N-type LD )   Rpd converts this reverse current into a feedback voltage.   The OPA's inverting input is connected to this feedback point. The OPA non-inverting input is connected to a 10-bit DAC which I can digitally control through the MCU.  Therefore, this DAC sets the peak LD current/optical power.  The OPA drives enough current through Q1's base in order to achieve the requested peak voltage at the Rpd point.

The LD I am using are all either IR or red LD's.   Their monitor photodiode current can varying wildly:  I've seen as low as 50uA all the way to 250uA for the same optical output power using LD's from the same box.   

Now in order to achieve varying optical powers, I utilize PWM and vary the duty cycle of a special feature of this OPA.   This particular OPA can switch its output pin within 50ns between (a) actively driven and (b) high-Z      When driving, the OPA servos current to Q1's base and ramps to/holds the peak laser power.  When in high-Z, Rbe removes base charge and quickly puts Q1 into cutoff.   

The Vref, DAC, and OPA are all actually integrated onto a single die along with the MCU itself.  For reference, it is a PIC16F17** series MCU.   The OPA has a 3.5MHz GBWP, 40 degree phase margin, 3V/us slew rate, a common-mode range of 0-VDD and an open-loop gain of 90dB.  I'm running this MCU and thus this OPA at 3.3V.   For reasons I cannot get into here, I am more or less forced to use an integrated approach like this rather than a separate OPA, MCU, DAC, etc.

About the feedback network:   I had to really wing this part of the design and just solder and unsolder caps until I got a somewhat workable response.  Cf1:  anything below 3.3nF and the base drive oscillates wildly.  I finally settled on Cf1 = 10nF and get somewhat stable results.   Cf2 and Rf are currently just placeholders, to be used if I can figure out how to properly do Type II compensation.   I am not sure how I would calculate this beforehand. 

Q1 is a 2SD1979GSL  NPN.   I have several other NPN's on hand I've been experimenting with.  The best results I've found are with NPNs that have a high hfe  and a lower ft range.   I tried originally using several MOSFETs in this scheme, but all of them horribly oscillated and I could not get anywhere near a flat response with any NMOS I tried here.   Probably had to do with the OPA's ability to drive a capacitive load.

Now, some scope images to illustrate my problem:



YELLOW trace is the feedback voltage at Rpd.  GREEN trace is the voltage on Q1's base.  PURPLE trace is the LD Anode ( the LD_SUPPLY )   RED trace is the LD cathode.   LIGHT PINK trace is  (LDA - LDK) for a differential measurement from LDA-LDK, giving me an idea of the LD forward voltage.   Here, I have the PWM frequency set to 4kHz and illustrated is a 5% duty cycle.  I start with the DAC set to something very small, like 100mV.  Then I adjust the DAC output upwards until I can trigger on the edge of the Rpd feedback signal in YELLOW. 

I see a rather fast ramp up with significant overshoot on the Rpd signal.  Not sure why that occurs here. 

Also, I see a rather fast ramp up to ~500mV on the base drive as soon as the PWM signal goes high and ... but then a slow and steady linearly increasing ramp of the base voltage until finally it flattens out.  It is this slow and steady ramp up of Vbe that hinders low duty cycles.   A zoomed in scope shot:



Upon the PWM signal going high/OPA driving, the base voltage shoots up quickly to about 500mV -- below threshold -- and then slooooooowwllly ramps up as you can see in the above scope shot.  This is a 5% duty cycle ON pulse shown.   If I go to 1% duty cycle, the entirety of the ON PWM signal is spent in this slow ramp-up, which keeps Vbe below threshold the entire time, effectively giving me zero output on the LD.    When this happens surprisingly, I still get a photocurrent feedback signal, which I don't understand at all.  If there is no lasing action, there should not be any feedback signal ...

I'm hoping to gather insight into the following points:

(1)  Is there anything in this circuit that strikes you as faulty or terribly incorrect?  Are any of my assumptions flawed or incorrect?
(2) Is there a better way to go about compensating this OPA for a better transient response and reduce the ringing on the rising edge?  Can I do this empirically as before?  I do not have access to a VNA to do fancy stuff like the loop gain/phase measurement ... 
(3) Why does the OPA seem to quickly slew to 500mV, but then exhibit that slow ramp-up to threshold?
(4) Can this integrated OPA do what I am asking of it, given its GBWP and slew rate?   
(5) Any other ways to improve this circuit by adding or deleted external elements?
(6) Would an NMOS work better here, if I were somehow able to get the circuit to not oscillate?
(7) Eventually I would like to get this circuit to work up to 10kHz modulation rates with 1% min duty cycles.  Would that be possible given this OPA? 

Thanks for reading! 

[janoc]

Well, this is a bit outside of my field of experience but I have a hunch that the opamp in the PIC is not designed to be abused like this. It is  a very slow opamp - 3V/us slew rate is not stellar and this will directly affect the current ramp-up. If I see correctly, your scope captures are at 20us/division, so that ramp-up seems to pretty much match the declared slew rate of your opamp.

The other thing is that as soon as you turn the laser on, the photodiode is doing negative feedback and is acting against the rise of the current.

The fast rise/overshoot may be spurious and due to your probing (parasitic inductance of the wires).


Also who knows how the opamp output behaves when you are turning the pin driver on and off (well, on and high-z). That's not normally the way PWM is done and the PIC pin drivers may not like that too much.

I would probably prototype this first by using a discrete opamp and a signal generator, just to make sure that the rest of the design isn't doing something weird before messing with the PIC.

[adamcord]

(3) Why does the OPA seem to quickly slew to 500mV, but then exhibit that slow ramp-up to threshold?

The way you have the OPA configured, with a feedback capacitor, is like an integrator, which is my guess as to why it shows that ramping behavior. Can you probe the output of the OPA before the 330 Ohm resistor? I would expect that to look like a ramp.

[StillTrying]

A nice long post, but too much for me to digest in one meal.

I don't think the first bit of your yellow PD signal is the PD, I think it's the op amp's output rising when enabled, and 'the ramp' is CF1 discharging. then the op amp 'works'.

The PWM plus DAC control seems complicated, especially as it's the opamp's output which is switched on and off.

 

[smoothVTer]

I think you're right about the photodiode ...

To make sure my LD/PD is still good, I took the LD/PD device out of its mount and into a LD controller to measure the photocurrent.  Then I forgot to put it back in circuit, and powered up everything with probes in place.  Surprisingly the yellow trace remains the same ... in the total absence of the photodiode current!  How does this make sense?  There is no DC path from OPA OUT  to OPA -IN  ... I don't understand how this could happen.

 

[jmelson]

Rpd is too high, by a couple orders of magnitude.  Instead of using a voltage amp, why not a transconductance amp.  There, the effective impedance is very low.
You will need a feedback resistor across the op-amp, value sets the conversion from current to voltage.
The rest of this circuit should work.  I built something pretty close to this some time ago, for writing images on photographic film with a laser diode.

Jon

[smoothVTer]

I'll try building up this circuit with a discreet single-supply OPA, with a GBWP somewhat higher than 20MHz.  I'll also use a potentiometer rather than a DAC.   Try to test out the concept with higher bandwidth components and work from there. Maybe it is something weird to do with the PIC's integrated OPA.

Regarding the toggling between driving and high-Z mode:    This is actually a feature of the chip built into the registers and hardware.  This PIC has the OPA "output override" function which can be internally routed to toggle based on the internally generated PWM signal.   Also there is at least one app note out there from Microchips website suggesting to use this high-z toggling functionality, albeit in an LED dimmer, DC/DC conversion sense.   

Using discreets, I can't toggle the OPA between driving and high-z because thus far I haven't found any OPA that advertises this.  If I can't toggle between high-z and driving mode, how else to achieve this variable duty cycle modulation?  I can't turn the OPA on and off fast enough with an EN pin ( OPA startup/shutdown time ).   I can't toggle an external DAC between 0V and some setpoint fast enough ( DAC settling time + cost of a fast settling DAC + size of an external DAC )   

[smoothVTer]

Rpd is too high, by a couple orders of magnitude.  Instead of using a voltage amp, why not a transconductance amp.  There, the effective impedance is very low.
You will need a feedback resistor across the op-amp, value sets the conversion from current to voltage.
The rest of this circuit should work.  I built something pretty close to this some time ago, for writing images on photographic film with a laser diode.

Jon

The nominal photocurrent at peak power output, measured by me, is 150uA.  But this can vary between 50uA to 250uA in the same part number laser diodes...

150uA * 2.49k = 0.374V

0.374V is about 1/3 of the headroom of the DAC's possible output ( Vref feeding the DAC is 1.024V )    So I think this resistor is more or less in the correct range, no?   Some room for adjustment down, more room for adjustment up. 

Also: you mean a transimpedance amplifier ( not transconductance ) ?   

If you meant transimpedance amplifier, there is a single reason I cannot use one: This is a single supply system, and as such, I cannot use a TIA because using a TIA with an N-type laser diode would generate a negative output voltage in proportion to Ipd * Rf      Since I don't have a negative supply, I cannot use a TIA in this fashion.

[janoc]

I'll try building up this circuit with a discreet single-supply OPA, with a GBWP somewhat higher than 20MHz.  I'll also use a potentiometer rather than a DAC.   Try to test out the concept with higher bandwidth components and work from there. Maybe it is something weird to do with the PIC's integrated OPA.

GBWP is not what matters so much, do make sure your opamp has sufficient slew rate.

Using discreets, I can't toggle the OPA between driving and high-z because thus far I haven't found any OPA that advertises this. 

What's wrong with using an extra transistor there? 

[StillTrying]

"If you meant transimpedance amplifier, there is a single reason I cannot use one: This is a single supply system, and as such, I cannot use a TIA because using a TIA with an N-type laser diode would generate a negative output voltage in proportion."

All you've got to do is leave out RPD, and provide a resistance path opamp output to -IN, the PD's current is already in the right phase for overall negative feedback.

[mikerj]

Just to make sure I understand your intentions, you want to provide a constant average power output independent of PWM duty cycle, so low duty cycles require much higher peak currents than high duty cycles?  I don't see any current limiting in the circuit, does the 2.5v laser supply guarantee you can't blow it up if the drive transistor is saturated?

[smoothVTer]

I'll try building up this circuit with a discreet single-supply OPA, with a GBWP somewhat higher than 20MHz.  I'll also use a potentiometer rather than a DAC.   Try to test out the concept with higher bandwidth components and work from there. Maybe it is something weird to do with the PIC's integrated OPA.

GBWP is not what matters so much, do make sure your opamp has sufficient slew rate.

Using discreets, I can't toggle the OPA between driving and high-z because thus far I haven't found any OPA that advertises this. 

What's wrong with using an extra transistor there? 

You're right ... I could have back-to-back NFETs at the OPA output pin to force the high-z state on any OPA I choose ... however ... at least in my preliminary experiments with this, I find large negative and positive going spikes upon the switch on/off state.  This might have to do with charge injection of the series mosfets, in which case I'd be worried about momentarily overdriving Q1's base and blowing the LD.  For example:



Here's switching the gate of the back-to-back NFETs and the signal transmission of a 1MHz sine wave:


Zoomed into the rising switching edge, I get some nasty 1V spikes positive on the ON edge and -1V spikes on the OFF edge:


This may or may not work ... I could choose much higher Ron NFETs to minimize the gate capacitance and thus the charge injection.

[smoothVTer]

Just to make sure I understand your intentions, you want to provide a constant average power output independent of PWM duty cycle, so low duty cycles require much higher peak currents than high duty cycles?  I don't see any current limiting in the circuit, does the 2.5v laser supply guarantee you can't blow it up if the drive transistor is saturated?

The idea is to be able to set the peak LD current to a constant, known level, independent of the LD's own Ipd characteristic.  Once the LD is "peaked", so to speak, the idea is to modulate at several kHz and vary the duty cycle to scale the output power.    The complication arises because the same part number laser diode ( red or IR, in this case )  can have an Ipd anywhere from 50uA to 250uA for an output power Po = 10mW     The LD is most efficient at this "peaked" output level, so to achieve highest efficiency, you want to always reach that peak power during the PWM "on" time.

I guess another way to word it is that I want to achieve a constant peak power, that I can then PWM/vary the duty cycle to scale the perceived average power.  To achieve this constant peak power, I run this system as a fast linear regulator, using the DAC as my peak power control signal.

[ajb]

There's a fundamental problem with switching the output of the opamp as you're doing: you're breaking the feedback loop.  An op amp without feedback is a comparator, and not a very good one, often exhibiting problematic behavior when changing state.  The fast rising edge, resultant ringing, and slow ramp up to the set point is entirely consistent with an op amp coming out of saturation and getting a hold of its feedback again.  You will almost certainly achieve better results by switching the setpoint on and off, and allowing the op amp to stay closed loop at all times. 

[duak]

I last worked with laser diodes in the early 90's.  I don't think things have changed much though: https://www.newport.com/medias/sys_master/images/images/he9/hd7/8797049520158/AN05-Laser-Diode-Characteristics-Overview.pdf  It explains some aspects but for what it's worth, here's my spin on it too:

- it's normal to have a variation in monitor current.  The photodiode sits at the other end of the laser diode chip and measures how much light leaks out of the cavity back facet or mirror. I can't remember the exact values for the leakage but you can see that a difference between 99.9% and 99.99% reflectivity is an order of magnitude variation in monitor current.
- a laser emits non-coherent radiation as an LED below lasing threshold.  The amount and wavelength is not usually specified.
- the LD_supply voltage seems a little low.  It should work but Q1 is running close to saturation ie., VBE > VCE
- you've got to be really careful not to exceed either the forward current limit or the power limit spec or you will blow the facet off the die and it will no longer lase.  This may have changed over the years, but back then the manufacturer wasn't kidding about exceeding limits.  We joked that you could kill a $100 device in 100 ns for a burn rate of $1B per second.
- back when I worked with these things, if the laser was modulated, the monitor was sampled and used to determine the drive current.  This was then switched or steered so as to not cause any current spikes that could damage the laser diode.

[mikerj]

Just to make sure I understand your intentions, you want to provide a constant average power output independent of PWM duty cycle, so low duty cycles require much higher peak currents than high duty cycles?  I don't see any current limiting in the circuit, does the 2.5v laser supply guarantee you can't blow it up if the drive transistor is saturated?

The idea is to be able to set the peak LD current to a constant, known level, independent of the LD's own Ipd characteristic.  Once the LD is "peaked", so to speak, the idea is to modulate at several kHz and vary the duty cycle to scale the output power.    The complication arises because the same part number laser diode ( red or IR, in this case )  can have an Ipd anywhere from 50uA to 250uA for an output power Po = 10mW     The LD is most efficient at this "peaked" output level, so to achieve highest efficiency, you want to always reach that peak power during the PWM "on" time.

I guess another way to word it is that I want to achieve a constant peak power, that I can then PWM/vary the duty cycle to scale the perceived average power.  To achieve this constant peak power, I run this system as a fast linear regulator, using the DAC as my peak power control signal.

Sounds to me like you want a constant current current drive.  Using a continuous feedback with an integrator to control laser current will result in the loop trying to maintain a constant average power, i.e. laser current will rapidly increase at low duty cycles.

If you want closed loop power control only whilst the laser is operating you will need some kind of sample and hold on the feedback signal gated by the PWM.  You may be able to do the entire power control loop by triggering an ADC conversion from the PWM output and using this to adjust the DAC which controls a current source.

[smoothVTer]

"If you meant transimpedance amplifier, there is a single reason I cannot use one: This is a single supply system, and as such, I cannot use a TIA because using a TIA with an N-type laser diode would generate a negative output voltage in proportion."

All you've got to do is leave out RPD, and provide a resistance path opamp output to -IN, the PD's current is already in the right phase for overall negative feedback.

OK, I tried this approach but I am not sure how to get it working correctly.    As a starting point, here is a simulation of the steady state system in its current form ( ignoring any PWMing of the LD )



The laser diode is modeled more or less with D1/D2/D3
B1 is a simplistic model of the photodiode:   below lasing threshold, the photocurrent is very low.  As it approaches lasing threshold ( ~20mA ) the photocurrent increases drastically, and thereafter, approximates a linear curve for increasing laser diode forward current.   C2/R4 are there for OPA stability.   With this simulation, I can finely control the forward current through D1/2/3 using the Vref_drive ( which would come from my internal DAC )


When I apply your suggestion to use the PD feedback current directly in a TIA approach, I cannot get a usable current through D1/2/3 anymore, no matter what value R4 I choose.  It seems to regulate, albeit to a very low current:



Does the 2nd schematic/simulation I have shown here reflect the general idea of your suggestion?

[smoothVTer]

I last worked with laser diodes in the early 90's.  I don't think things have changed much though: https://www.newport.com/medias/sys_master/images/images/he9/hd7/8797049520158/AN05-Laser-Diode-Characteristics-Overview.pdf  It explains some aspects but for what it's worth, here's my spin on it too:

- it's normal to have a variation in monitor current.  The photodiode sits at the other end of the laser diode chip and measures how much light leaks out of the cavity back facet or mirror. I can't remember the exact values for the leakage but you can see that a difference between 99.9% and 99.99% reflectivity is an order of magnitude variation in monitor current.
- a laser emits non-coherent radiation as an LED below lasing threshold.  The amount and wavelength is not usually specified.
- the LD_supply voltage seems a little low.  It should work but Q1 is running close to saturation ie., VBE > VCE
- you've got to be really careful not to exceed either the forward current limit or the power limit spec or you will blow the facet off the die and it will no longer lase.  This may have changed over the years, but back then the manufacturer wasn't kidding about exceeding limits.  We joked that you could kill a $100 device in 100 ns for a burn rate of $1B per second.
- back when I worked with these things, if the laser was modulated, the monitor was sampled and used to determine the drive current.  This was then switched or steered so as to not cause any current spikes that could damage the laser diode.

Thanks for the resources and your inputs.

The devices I am working on use very cheap laser diodes ( 635nm reds and 850nm IR's ) that cost about $3/ea. so even if I blow several it is not a big deal ( phew! )     The system itself is tiny and runs off a small, weak battery.  Therefore the smallest PCB I can get away with steers me towards a fully integrated solution, if I can get away with it. Also having these weak batteries makes it critical to drive these LD's at peak efficiency to get the longest battery life. 

[smoothVTer]

Just to make sure I understand your intentions, you want to provide a constant average power output independent of PWM duty cycle, so low duty cycles require much higher peak currents than high duty cycles?  I don't see any current limiting in the circuit, does the 2.5v laser supply guarantee you can't blow it up if the drive transistor is saturated?

The idea is to be able to set the peak LD current to a constant, known level, independent of the LD's own Ipd characteristic.  Once the LD is "peaked", so to speak, the idea is to modulate at several kHz and vary the duty cycle to scale the output power.    The complication arises because the same part number laser diode ( red or IR, in this case )  can have an Ipd anywhere from 50uA to 250uA for an output power Po = 10mW     The LD is most efficient at this "peaked" output level, so to achieve highest efficiency, you want to always reach that peak power during the PWM "on" time.

I guess another way to word it is that I want to achieve a constant peak power, that I can then PWM/vary the duty cycle to scale the perceived average power.  To achieve this constant peak power, I run this system as a fast linear regulator, using the DAC as my peak power control signal.

Sounds to me like you want a constant current current drive.  Using a continuous feedback with an integrator to control laser current will result in the loop trying to maintain a constant average power, i.e. laser current will rapidly increase at low duty cycles.

If you want closed loop power control only whilst the laser is operating you will need some kind of sample and hold on the feedback signal gated by the PWM.  You may be able to do the entire power control loop by triggering an ADC conversion from the PWM output and using this to adjust the DAC which controls a current source.

Thanks mikerj,

I found this app note from Maxim about doing pretty much what you describe in the first half of your reply, I think.  Yet I do not understand how this circuit could possibly work ( regulating average power ) because then there is no limit to how much the current can increase at low duty cycles ... this circuit below would push the current very high and blow the laser diode at low duty cycles.

Taken from https://www.maximintegrated.com/en/app-notes/index.mvp/id/1811

[pwlps]

The theory of operation:   When a current flows through the LD, a reverse current proportional to the light output is generated through the PD.  The LD and PD are in the same 3-pin package ( this being an N-type LD )   Rpd converts this reverse current into a feedback voltage.   The OPA's inverting input is connected to this feedback point. The OPA non-inverting input is connected to a 10-bit DAC which I can digitally control through the MCU.  Therefore, this DAC sets the peak LD current/optical power.  The OPA drives enough current through Q1's base in order to achieve the requested peak voltage at the Rpd point.

The LD I am using are all either IR or red LD's.   Their monitor photodiode current can varying wildly:  I've seen as low as 50uA all the way to 250uA for the same optical output power using LD's from the same box.   

Now in order to achieve varying optical powers, I utilize PWM and vary the duty cycle of a special feature of this OPA.   This particular OPA can switch its output pin within 50ns between (a) actively driven and (b) high-Z      When driving, the OPA servos current to Q1's base and ramps to/holds the peak laser power.  When in high-Z, Rbe removes base charge and quickly puts Q1 into cutoff.   

 I don't understand the principle of operation as you explained, but I admit I didn't make much effort to analyze it in details.   
I don't get:  why should you need any digital stages if your pwm duty cycle is only controlled by an analog input ?
You might be interested to compare it with my approach here:
https://www.eevblog.com/forum/projects/simple-lifi-transmitting-audio-signal-with-a-led/msg2282771/#msg2282771

A slight modification would be needed if you want to have the feedback go through a measuring photodiode rather than to be taken directly from the LED output voltage, otherwise this sigma-delta converter circuit can give a high-fidelity PWM without any digital components. 

[jmelson]

0.374V is about 1/3 of the headroom of the DAC's possible output ( Vref feeding the DAC is 1.024V )    So I think this resistor is more or less in the correct range, no?   Some room for adjustment down, more room for adjustment up. 

Yes, in a strictly DC sense, but your problem is poor AC bandwidth.  High impedances wreck your bandwidth.

Also: you mean a transimpedance amplifier ( not transconductance ) ?   

If you meant transimpedance amplifier, there is a single reason I cannot use one: This is a single supply system, and as such, I cannot use a TIA because using a TIA with an N-type laser diode would generate a negative output voltage in proportion to Ipd * Rf      Since I don't have a negative supply, I cannot use a TIA in this fashion.

Yes, right, incorrect term, I did mean transimpedance.  Well, if the output is at the wrong DC resting point, then you have to shift it.  Since you don't have a negative rail to bias the input, you need to shift the non-inverting input more positive.  When set right, that will cause the output to idle at a more positive value, and then move in the minus direction with increasing laser output.  But, this presents a low impedance to the photodiode, and will therefore improve the AC response.

Jon

[chris_leyson]

Hi smoothVTer, it's been a few years since I last worked with laser diodes and back then we were driving them CW.
I probably had a mini rant about fixing some of the legacy laser drivers that we used because 50% current overshoot was not uncommon, all down to putting too many 100n feedback caps around op-amps and then not testing the transient response  Any new laser drivers, ~50mW power, just used a single op-amp driving a mosfet but needed to have a select on test resistor to set the correct PD feedback voltage for each laser diode. I used just enough capacitance to roll off the high frequency gain and managed to get a good step response with 5% overshhot. The only problem with a single op-amp solution was that if you have a set point of zero then the laser diode is always driven at it's threshold current.
I was looking at fast(ish) laser drivers for a personal project but the company I worked went belly up so that never got finished, I had more important things to think about.

Anyway, something you might find useful is a little circuit to mimic or emulate a small laser diode with an N style pinout.
The laser diode drive transistor or mosfet is represented by the current source I1 and diodes D1 and D2 together with Q1 Vbe mimic the LD forward voltage somewhat. R1 sets the threshold current for the emulator, 30 ohms gives about 20mA, Q1 and R2 drive an LED at about 10mA for a max LD drive current of 50mA in this example and Q2 and R3 mimic the PD feedback current set to 330uA at 50mA drive.  It's a reasonable approximation to a small laser diode and a lot cheaper to fix if you let the magic smoke out.

[StillTrying]

I don't think any of these solution work very well, so here's yet another.

[mikerj]

Thanks mikerj,

I found this app note from Maxim about doing pretty much what you describe in the first half of your reply, I think.  Yet I do not understand how this circuit could possibly work ( regulating average power ) because then there is no limit to how much the current can increase at low duty cycles ... this circuit below would push the current very high and blow the laser diode at low duty cycles.

Taken from https://www.maximintegrated.com/en/app-notes/index.mvp/id/1811

When using a laser for transmitting data the data signal is always DC balanced, i.e. the average duty cycle (over some number of bits) is maintained at 50% i.e. an equal number of ones and zeros for an NRZ signal.  Not only does this maintain a fixed average power output, it makes the receivers job far easier as the slicing level can be set based on average power.

That app note does raise another issue that you might run into; biasing the laser.  For the fastest response you don't want to turn the laser fully off as this slows the response considerably.  Typically the laser will be driven so that it always remains above it's threshold current, though at kHz speeds this may not be an issue.

[awallin]

not sure if it's been said/linked already but the usual constant-current circuit used for sciency-stuff is:
Libbrecth-Hall http://www.submm.caltech.edu/kids_html/DesignLog/DesignLog179/MillerMUSICReadoutDocs/HEMT%20Power%20Supply/Libbrecht%20and%20Hall,%20A%20Low%20Noise%20High%20Speed%20Diode%20Laser%20Current%20Controller.pdf
Erickson https://arxiv.org/abs/0805.0015
Seck https://arxiv.org/abs/1604.00374